Engineers waste
a year reading PDF datasheets.
3 in 4 engineers lose at least half a day every week reading datasheets.

* Estimate, range $10–91B · survey Accuris / Fuld & Co., n=439, 2026
www.make-it.tech01The datasheet never went digital.
Agents read datasheets now. Datasheets are PDFs made for humans. AI agents need data.
www.make-it.tech02
Living Datasheets that answer back.
Data for Humans and Agents
IRFZ44NPbF datasheet © Infineon Technologies AG, used for demonstration.
www.make-it.tech03Point at any graph. Get the numbers in seconds.
Engineers point at a curve and get the value, with its source page.
Turns every datasheet PDF in the world into searchable, cited data.
IRFZ44NPbF datasheet © Infineon Technologies AG, used for demonstration.
www.make-it.tech04The datasheet says
49 Amps
In the real world
about 8 Amps
at your operating point:
free air, no heatsink
Found in minutes, before you build the board.
IRFZ44NPbF datasheet © Infineon Technologies AG, used for demonstration.
www.make-it.tech05Built at the hackathon,
www.make-it.tech06From datasheet PDF to cited answers
Measured: cost on 63 PDFs, extraction only (grok-4.5, median); search on 21 questions, 9 datasheets. Lilac = AI model call.
www.make-it.tech07Not another tool.
One source of truth.
The data layer between chipmakers, engineers and their AI agents.
Every value verified. Every value cited, on the engineer’s screen and in the agent’s answer.
FLYWHEEL More datasheets → more engineers & agents → more usage data → better datasheets
same data · same citation verified in · orders outKiCad plugin, web app, browser extension
Structured API, RAG, MCP
for the AI era
Publish verified data, see real demand
Spec-based search, direct purchase
www.make-it.tech0801 Open source
Open core, available to the community, to grow the user base exponentially.
🌱02 First customers
Turn the engineers and teams who use it into our first customers.
🌳03 B2B
Grow into a B2B offer for big companies and manufacturers.
🌟Engineers already pay for their tools.
62,768
Altium subscribers
AUD 9.1B
Renesas bought Altium
www.make-it.tech09We build hardware for a living. We feel this pain every week.

Francisco Mendes
CEO · product & pitch

Pedro Magalhães
CTO · architecture

Miguel Tavares
EE · extraction pipeline

Vitor Mendes
Software · AI & query layer
www.make-it.tech10SpaceX.ai »»
“Help us promote this open source in the community. I’m sure SpaceX hardware engineers are spending too much time reading PDFs!”
this deck, live:make-it.tech/virtual-datasheet
Stop reading datasheets with a ruler,

www.make-it.tech11On a real datasheet: IRFZ44NPbF © Infineon Technologies AG, used for demonstration.
www.make-it.techA1Provenance metadata
Every value is stored with {page, bbox, extractor, confidence, review_status}. Click a number and its printed source is highlighted on the original PDF.
Curve digitization
OpenCV finds the plot frame, calibrates lin/log axes and tracks each curve. PCHIP interpolation lets you query any operating point, not just the printed grid.
Normalized schema
JSON schema v2, SI units, canonical keys and test conditions (TC, VGS…) per value. Validated with jsonschema; low-confidence rows flagged needs_review.
Embeddings & hybrid search
Datasheet statements plus application context → text-embedding-3-small (1536-d) → pgvector, cosine distance. Numbers stay in SQL filters; vectors only for meaning.
LLM with guardrails
Grok (xAI API) extracts only where rules find no table. Each value must quote text found on the page, or it is marked needs_review. No free-form LLM numbers.
Design context
KiCad project → operating envelope (supply, load current, cooling, gate drive, PWM). A deterministic rule engine gives pass/fail with margins and ranks alternatives.
Stack: Python 3.11 · FastAPI · Postgres 16 + pgvector (Docker) · PyMuPDF · pdfplumber · OpenCV · NumPy · PDF.js · D3 · pytest · Playwright · Cursor agents + GitHub PRs
Cited AI chat (answers must cite parameter, page and bbox; uncited claims refused): API contract defined. MCP server: roadmap.
www.make-it.techA2The core features that already save engineers time, on real datasheets.
Every part, every vendor, every value cited back to its page.
For humans and for their AI agents.
■ today: 253 datasheets digitized (1 icon ≈ 10) · □ next: all of them
www.make-it.techA3